Remnant Fieldworks

The Public Record

RF Research Index

The RF Research Index is the documented experimental record of Remnant Fieldworks Inc.’s research program. Experiments are published to the RF Zenodo community with DOIs assigned per publication.

Corpus summary (as of 2026-10-02)

108 documented experiments across 13 research families: 96 PASS, 9 preserved FAIL, and 3 special-status outcomes. The public record includes PASS, preserved FAIL, remediation, and qualified or special-status outcomes.

Publication record

Research is published to the RF Zenodo community:

https://zenodo.org/communities/remnant-fieldworks/

Code and supporting materials for the research program are published across public GitHub repositories under github.com/derekhone, with availability varying by experiment family.

CUDA-EP

Real GPU execution: failure preserved, remediation retested

We put an ExecutionProof gateway in front of a real CUDA kernel launch on an NVIDIA Tesla T4. CUDA-EP-001 was preregistered and executed on real hardware. The exclusive-launcher control held, the functional reference vectors matched, and NVIDIA Compute Sanitizer ran clean. But one preregistered control case returned ALLOW where we expected HOLD, so the experiment is published as a partial failure rather than rewritten as a success.

We then created a separate preregistered remediation, CUDA-EP-001R1. The CUDA kernel and launcher were carried over byte-identical while the control-state, ProofRecord-verification, and compute-capability checks were corrected. The retest passed all five targeted cases, including the exclusive-launcher control.

Both records are public and separately preserved:

This is a bounded result on one class of NVIDIA GPU. It does not claim generalized CUDA security, production readiness, independent validation, or NVIDIA endorsement. The value of the record is that the failure, remediation, and retest are all inspectable.

The 13 experiment families (108 experiments total)

SeriesExperiments
ARK (441-503)65
WITNESS (1-4)4
BELLWETHER (1-3)3
CHRONO-11
OMNI-11
TRINITY (1-2)2
DM (001-005)5
FIDELITY (IF-01-06)6
QG (001-004)4
EP-SEC (001-009b)10
IB (001-003)3
AUTH (001-002)2
CUDA-EP (001, 001R1)2
13-family total108

Coherent Inheritance Framework (CIF)

The Coherent Inheritance Framework provides a theoretical framework used across several RF research programs. It is documented in the CIF preprint (Zenodo DOI: 10.5281/zenodo.21753238) and a companion ancient systems case study series (DOI: 10.5281/zenodo.21716831). CIF is the framework; the experiment families above test and apply its principles. The CIF preprint and case study corpus are not counted within the 108-experiment figure.

CIF-LAAD (separate candidate study)

CIF-LAAD (Coherent Inheritance Framework for Low-Altitude Air Domain Awareness) is a separate candidate study applying CIF principles to sensor fusion falsification in the low-altitude air domain, with relevance to air-defense applications. It is a software-based simulation at Technology Readiness Level 3. It is not a hardware experiment and has not undergone independent validation. Two Zenodo records are associated with this work:

  • Stage 0 falsification study: DOI 10.5281/zenodo.22242729 - “CIF-LAAD Stage 0: Coherence-Gated Inheritance for Low-Altitude Air Defense Tracking - Falsification Study and Batch-Mode Observation”
  • Characterization record: DOI 10.5281/zenodo.22262862 - “CIF-LAAD: Coherent Inheritance Framework for Low-Altitude Air Domain Awareness, Characterization Record”

CIF-LAAD is maintained as a distinct research program and is not counted within the 108-experiment figure.

Additional programs

The following programs are documented in the public record but are not counted within the 108-experiment corpus:

  • CompositionProof - tests multi-witness simultaneous composition certification. Code: compositionproof-testbeds. Zenodo DOI: 10.5281/zenodo.22666465.
  • QT-001 (Quantum Trust Program) - a preregistered quantum calibration study. Design and preregistration are SHA-locked. Hardware execution is pending QPU allocation. No results reported to date.

Methodology notes

  • -Quantum experiments (WITNESS, BELLWETHER, CHRONO, OMNI, TRINITY, QG) use IBM Quantum hardware unless noted otherwise in the individual publication.
  • -DM series experiments use quantum simulation methods.
  • -FIDELITY and IB series use classical inference applied to agent behavior.
  • -EP-SEC series uses adversarial test methods against the execution boundary.
  • -CIF-LAAD Stage 0 is software simulation only (TRL 3). No hardware. No independent validation.